ViZDoom test – visual Reinforcement Learning (health gathering supreme)

Agent trained in ViZDoom environment (http://vizdoom.cs.put.edu.pl/) using visual input, current health points and time. The learning algorithm was based on DeepMind’s DQN.

Agent is placed in an acidic environment and has to collect medikits and avoid blue poison vials to survive.

Post Author: hatefull